First Tuesday of the month, 9:14 a.m.: a twelve-page PDF lands in a home-services owner’s inbox. He pays $4,000 a month for someone to manage his Google Ads account, and this is what he can see of the work.
This is a composite, drawn from too many versions of the same call to pretend it was one owner. Say he spends around $20,000 a month on ads. Say he opens the PDF while his coffee cools. Logo on the cover. Impressions up. Clicks up. Click-through rate up, with a little green arrow beside it. There is a bar chart, a pie chart, and a screenshot of an ad with a good Quality Score. He scrolls to the end and thinks what he thought last month: this looks fine.
His cost per booked job has crept up 11% since spring. Nobody put that sentence on the cover.
I know the shape of that PDF. I have seen a hundred versions from a dozen shops. Impressions, clicks, CTR: numbers that fit neatly in a chart and make the meeting easier to finish. The fee does not surprise me, either. For a mid-market account spending $15,000 to $20,000 a month, $1,500 to $4,000 in monthly management fees sits within the usual band. Four thousand dollars is not a scam price. It is a confident price. It implies someone is paying close attention between reports.
The report measures attention paid to the ads, not attention paid to the account. Clicks and CTR tell you whether people noticed the ad. They do not tell you whether those people booked a job. CTR can rise while lead volume falls, and in this composite, brand searches help lift the headline number while the cost of a booked job keeps climbing. The business is paying more to get the outcome it actually needs, but the outcome is buried on page nine without a comparison column.
The owner is not missing a secret formula. He is reading the document he was given. A green arrow next to CTR asks him to feel reassured; a cost-per-job number stranded in a table asks him to do the agency’s analysis for it. That is a tidy arrangement for the report. It is a poor one for the person paying the invoices.
He asks for a closer look. I ask for view-only access, as I do when someone says an account has been flat for six months. In Google Ads, I open Tools, then Change History, and set the range to 180 days. An agency can choose which chart goes into a PDF. Change History is less accommodating: work appears there with a timestamp. Long gaps appear there, too.
Eleven changes in six months. Two bid adjustments. One ad pause. A budget bump in June. The history does not prove that nobody thought about the account, and it does not tell us what every conversation sounded like. It does show how little changed inside it while the business kept spending. The gaps between those eleven changes are the story.
The search terms tab gives those gaps a price. Close variants have piled up for months: jobs and hiring queries, DIY how-to searches, a competitor’s accessory brand that never buys anything, and three suburbs the owner does not service. People are searching, the ads are appearing, and money is leaving the account. The wrong searches do not arrive all at once with an alarm attached. They arrive a few at a time, week after week, until the total looks ordinary.
I used to mine search terms at 1 a.m., adding negatives in batches because broad match will spend wherever you let it. It is mechanical work, but that does not make it optional. In this account, nobody has added the negatives that would stop those particular clicks. Next week, the same searches can come back. The report can still say CTR is up.
Then there is the June budget change. Someone raises daily caps 30% on a Friday afternoon. Two Smart Bidding campaigns go back into learning before a holiday weekend, and the change sits unmonitored while the campaigns retune. A learning period after a change is not, by itself, proof of bad management. The question is whether anyone watches what happens next. This explanation of the learning phase separates a real calibration window from an excuse for leaving an account alone. Here, the budget moves and the follow-up does not.
The account learns on his money while no one is in the room.
The losses are not dramatic on any one day. In February, the wrong search terms cost him maybe $800. By May, the same leak costs $1,400 as bids creep up and broad match finds more places to spend. The June change adds another stretch of expensive clicks. Each week can be explained away as a fluctuation; together, the weeks explain why the budget keeps running while booked jobs fail to grow. That is the trouble with looking only when a report is due. A month of small, fixable losses becomes one number after the chances to prevent them have passed.
By August, I put the six-month math on a screen where the owner can see it. Roughly $20,000 in ad spend plus $4,000 in management fees means $24,000 out the door each month. Over six months, that is $144,000. His booked jobs have not grown, and his cost per booked job is up roughly a tenth. The report calls the account stable. He has paid a top-of-band fee for a business outcome moving the wrong way.
That does not mean the whole $144,000 was wasted. Ads brought in jobs. Management fees paid for some work, too; the change log is sparse, not blank. The point of putting the numbers together is to see the scale of the decision. When someone says a slow month is not worth worrying about, it helps to remember what another month costs. The owner does not need a prettier explanation of impressions. He needs to know what will change before the next $24,000 goes through.
On the call, he asks the question straight: six months flat, $4,000 a month for management—is AI the better option now?
For this account, I say yes. Not because a machine has discovered a cleverer ad headline. The tactics we are talking about are plain: check search terms, exclude waste, watch the budget change, and measure booked jobs rather than admiring clicks. The problem is that those tasks wait for a person’s next review while auctions and queries keep moving. I can be a good manager for the hours I am inside an account. The account still runs when I close the laptop.
An AI agent for Google Ads earns its keep in those hours. It can keep working through search terms, budget shifts, and performance changes rather than waiting for the monthly PDF to make them visible. That does not make every automated action a good one. It means the work can happen close to the moment when there is still something to do about the problem.
I used to tell clients attention was the product. I was wrong about who had to supply all of it. Periodic attention is the failure mode, not the fee. Four thousand dollars for attentive management would be a different story. Four thousand dollars for a report on what drifted while nobody watched is this one.
I do not tell him to fire the agency on the call. Switching managers has a cost, and anyone who skips that part is selling something. A transition can take weeks while tracking is checked, changes are made, and Smart Bidding retunes; performance can dip during that handover. I have lived through that dip from the agency side. If an account has only recently stalled after improving, I would not treat one flat report as a verdict. Six months of flat booked jobs, rising cost per job, and little visible work is a different pattern.
I also tell him who should skip the AI move. If a business spends $3,000 a month and changes its offer every quarter, as that chaotic SaaS startup I used to run ads for did, it may not give a system a stable signal to learn from. If tracking counts every form reload as a lead, automation can chase that bad signal faster than a person. And if the job is to attend a Monday meeting and argue about brand colors, hire the agency. What the deck calls synergy, I call two people doing the same job. An autonomous engine will not attend the meeting. It will read the search terms at 2 a.m.
The owner asks what I would check before deciding. I give him three screens, in order:
His screens tell the same story from three angles. Few changes. Search terms that should have been excluded. Reporting that looks better than the booked-job cost. groas has a fuller list of agency red flags, but he does not need another list to understand what is on his screen. He laughs when I read the reporting one aloud. He has six copies of it in his inbox.
He does not fire the agency that Friday. He puts it on a 30-day clock and gives the account to a system that works nights. In this composite, the change log begins to look like a worked account: search terms trimmed, budget moved out of the two zip codes that never book and into the four that do, brand terms separated from the general pool so their strong CTR can no longer flatter the prospecting numbers. By the second week, the $300 to $400 a week going to hiring queries and DIY traffic thins to almost nothing. Fewer bad auctions entered; more of the same budget is available for searches from people who need a truck to show up.
The first Tuesday of the next month arrives. At 9:14 a.m., there is no twelve-page PDF, no pie chart, no green arrow asking him to feel better about clicks. There is a short log of what changed that week and what happened to booked-job cost afterward. He reads it in his kitchen while the coffee is still hot.
He keeps one of the old PDFs in a drawer under the parts catalogs. It is the April one, the cleanest-looking of the six. Green arrows on the cover. Cost per job climbing on page nine.
